[Gllug] Algorithm question for your Friday morning

[Adrian McMenamin]

On Fri, November 17, 2006 11:24 am, Tom Schutzer-Weissmann wrote:
> On Fri, 2006-11-17 at 10:23 +0000, Adrian McMenamin wrote:
>
>> I have some code that currently takes an arithmetic mean of people's
>> preferences/votes. But it occurs to me that it would be better to
>> replace
>> that with some sort of algorithm that weights historical votes less than
>> more recent votes.
>>
>> (I believe top(1) uses an exponential decay function)
>>
>> Is there a simple way to do this when the only record I currently have
>> of
>> historic votes is the arithmetic mean?
>
> Morning Adrian,
>
> If you only have the mean value, then you don't have any record of
> historic votes and you won't be able to recreate them in order to apply
> a different function!?
>
> Here's a mean value: 4, average of only 2 votes. How many solutions to
> (a + b)/2 = 4 can you find on a Friday morning?
>
> Or do you mean, you have a _series_ of mean values, from the time when
> only one person had voted, through to the present? Or something?


Well I was hoping for an engineering heuristic :) - and I do have some
recorded means, yes: one for every hour for the last 20, one every day for
last 20, and similarly for week and month

-------------- next part --------------
-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug